Blind Deconvolution of Vehicle Inductance Signatures for Travel-Time Estimation

Abstract

Travel-time data provides vital information for traffic monitoring, management, and planning. The objective of this research was to develop a new computational approach that could accurately measure travel time from two sets of spatially separated loop detectors using re-identification of vehicle inductance signatures generated by the loops. Although measuring travel time using loop inductance signatures is not new, all past approaches essentially relied on pattern matching of raw inductance waveforms without restoring the loss of detailed features caused by a large detection zone of inductive loops. The main effort in this research was to develop a new computational algorithm that restores the lost details from the raw inductance waveforms by modeling the output of loop detectors as a convolution of the original vehicle signature and the loop system function. This restoration problem was formulated as a blind deconvolution problem since we know neither the impulse response of the loop detectors, nor the original vehicle signature. To solve this blind problem, two basic blind deconvolution approaches were used, Godard deconvolution and constrained least squares. Experimental results showed that both methods performed well and significantly exposed the original signature information with unique vehicle characteristics.